tpu-inference 0.11.1__py3-none-any.whl

tests/kernels/ragged_paged_attention_kernel_v3_test.py

@@ -0,0 +1,504 @@
+import jax
+import jax.numpy as jnp
+import numpy as np
+from absl.testing import absltest, parameterized
+from jax._src import dtypes
+from jax._src import test_util as jtu
+
+from tpu_inference.kernels.ragged_paged_attention.v3.kernel import (
+    ragged_paged_attention, ref_ragged_paged_attention)
+from tpu_inference.kernels.ragged_paged_attention.v3.util import (
+    align_to, cdiv, get_dtype_packing)
+
+jax.config.parse_flags_with_absl()
+
+
+@jtu.with_config(jax_numpy_dtype_promotion="standard")
+class RaggedPagedAttentionKernelTest(jtu.JaxTestCase):
+
+    def _test_ragged_paged_attention(
+        self,
+        seq_lens,  # List[(q_len, kv_len)]
+        num_heads,  # [num_q_heads, num_kv_heads]
+        head_dim,
+        page_size,
+        q_dtype,
+        kv_dtype,
+        num_pages,
+        *,
+        num_kv_pages_per_block=8,
+        num_queries_per_block=64,
+        vmem_limit_bytes=100 * 1024 * 1024,
+        max_num_batched_tokens=512,
+        max_num_seq=8,
+        sliding_window: int | None = None,
+        soft_cap: float | None = None,
+        q_scale: float | None = None,
+        k_scale: float | None = None,
+        v_scale: float | None = None,
+    ):
+        rng = np.random.default_rng(1234)
+
+        def gen_random(shape, dtype):
+            return jnp.array(rng.random(size=shape,
+                                        dtype=np.float32)).astype(dtype)
+
+        if not jtu.is_device_tpu_at_least(version=4):
+            self.skipTest("Expect TPUv4+")
+        cu_q_lens = [0]
+        kv_lens = []
+        for q_len, kv_len in seq_lens:
+            assert q_len <= kv_len
+            cu_q_lens.append(cu_q_lens[-1] + q_len)
+            kv_lens.append(kv_len)
+
+        max_num_batched_tokens = max(align_to(cu_q_lens[-1], 128),
+                                     max_num_batched_tokens)
+        max_num_seq = max(align_to(len(seq_lens), 8), max_num_seq)
+        max_kv_len = max(kv_lens)
+        pages_per_seq = cdiv(max_kv_len, page_size)
+        num_q_heads, num_kv_heads = num_heads
+
+        q = gen_random((max_num_batched_tokens, num_q_heads, head_dim),
+                       q_dtype)
+        k = gen_random((max_num_batched_tokens, num_kv_heads, head_dim),
+                       kv_dtype)
+        v = gen_random((max_num_batched_tokens, num_kv_heads, head_dim),
+                       kv_dtype)
+        page_cnt = 0
+        page_indices_list = []
+        kv_pages_list = []
+        kv_packing = get_dtype_packing(kv_dtype)
+        padded_head_dim = align_to(head_dim, 128)
+        num_kv_heads_x2 = align_to(num_kv_heads * 2, kv_packing)
+        for kv_len in kv_lens:
+            kv = gen_random((
+                kv_len,
+                num_kv_heads_x2 // kv_packing,
+                kv_packing,
+                padded_head_dim,
+            ), kv_dtype)
+            kv = jnp.pad(
+                kv,
+                (
+                    (
+                        0,
+                        cdiv(kv_len, page_size) * page_size - kv_len,
+                    ),
+                    (0, 0),
+                    (0, 0),
+                    (0, 0),
+                ),
+                constant_values=jnp.nan,
+            ).reshape(
+                -1,
+                page_size,
+                num_kv_heads_x2 // kv_packing,
+                kv_packing,
+                padded_head_dim,
+            )
+            indices = page_cnt + jnp.arange(kv.shape[0], dtype=jnp.int32)
+            indices = jnp.pad(
+                indices,
+                ((0, pages_per_seq - indices.shape[0]), ),
+                constant_values=jnp.nan,
+            )
+            page_indices_list.append(indices)
+            page_cnt += kv.shape[0]
+            kv_pages_list.append(kv)
+
+        kv_cache = jnp.concatenate(kv_pages_list, axis=0)
+        kv_cache = jnp.pad(
+            kv_cache,
+            ((0, num_pages - kv_cache.shape[0]), (0, 0), (0, 0), (0, 0),
+             (0, 0)),
+            constant_values=jnp.nan,
+        )
+        page_indices = jnp.stack(page_indices_list, axis=0)
+        page_indices = jnp.pad(
+            page_indices,
+            ((0, max_num_seq - page_indices.shape[0]), (0, 0)),
+            constant_values=jnp.nan,
+        )
+        page_indices = page_indices.reshape(-1)
+
+        cu_q_lens = jnp.array(cu_q_lens, dtype=jnp.int32)
+        cu_q_lens = jnp.pad(cu_q_lens,
+                            (0, max_num_seq + 1 - cu_q_lens.shape[0]))
+        kv_lens = jnp.array(kv_lens, dtype=jnp.int32)
+        kv_lens = jnp.pad(kv_lens, (0, max_num_seq - kv_lens.shape[0]))
+        distribution = jnp.array([0, 0, len(seq_lens)], dtype=jnp.int32)
+
+        args = (
+            q,
+            k,
+            v,
+            kv_cache,
+            kv_lens,
+            page_indices,
+            cu_q_lens,
+            distribution,
+        )
+
+        kwargs = {
+            "sliding_window": sliding_window,
+            "soft_cap": soft_cap,
+            "q_scale": q_scale,
+            "k_scale": k_scale,
+            "v_scale": v_scale,
+        }
+
+        expected, expected_kv_cache = ref_ragged_paged_attention(
+            *args,
+            **kwargs,
+        )
+
+        output, updated_kv_cache = ragged_paged_attention(
+            *args,
+            **kwargs,
+            num_kv_pages_per_block=num_kv_pages_per_block,
+            num_queries_per_block=num_queries_per_block,
+            vmem_limit_bytes=vmem_limit_bytes,
+        )
+        output = output[:cu_q_lens[distribution[-1]]]
+
+        dtype_bits = dtypes.bit_width(jnp.dtype(kv_dtype))
+        tols = {
+            32: 0.15,
+            16: 0.2,
+            8: 0.2,
+            4: 0.2,
+        }
+        tol = tols[dtype_bits]
+        self.assertAllClose(output, expected, atol=tol, rtol=tol)
+        mask = ~jnp.isnan(expected_kv_cache)
+        self.assertArraysEqual(updated_kv_cache[mask], expected_kv_cache[mask])
+        self.assertEqual(output.shape[-1], head_dim)
+
+    @parameterized.product(dtype=[jnp.float32, jnp.bfloat16], )
+    def test_ragged_paged_attention_basic(self, dtype):
+        seq_lens = [(192, 328), (128, 180), (64, 255)]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            dtype,
+            dtype,
+            num_pages,
+        )
+
+    # TODO: support integer (int8, int4) and fp4 kv cache
+    @parameterized.product(
+        q_dtype=[jnp.bfloat16],
+        kv_dtype=[jnp.float8_e5m2, jnp.float8_e4m3fn],
+        kv_scales=[(0.5, 0.5), (1.0, 1.0)],
+    )
+    def test_ragged_paged_attention_quantized_kv_cache(self, q_dtype, kv_dtype,
+                                                       kv_scales):
+        if not jtu.is_device_tpu_at_least(version=5):
+            self.skipTest("Expect TPUv5+")
+        seq_lens = [(192, 328), (128, 180), (64, 255)]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+        k_scale, v_scale = kv_scales
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            q_dtype,
+            kv_dtype,
+            num_pages,
+            k_scale=k_scale,
+            v_scale=v_scale,
+        )
+
+    @parameterized.product(
+        q_dtype=[jnp.bfloat16],
+        kv_dtype=[jnp.float8_e5m2, jnp.float8_e4m3fn],
+        q_scale=[0.5, 1.0],
+        kv_scales=[(0.5, 0.5), (1.0, 1.0)],
+    )
+    def test_ragged_paged_attention_quantized_attention(
+            self, q_dtype, kv_dtype, q_scale, kv_scales):
+        if not jtu.is_device_tpu_at_least(version=5):
+            self.skipTest("Expect TPUv5+")
+        seq_lens = [(192, 328), (128, 180), (64, 255)]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+        k_scale, v_scale = kv_scales
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            q_dtype,
+            kv_dtype,
+            num_pages,
+            q_scale=q_scale,
+            k_scale=k_scale,
+            v_scale=v_scale,
+        )
+
+    @parameterized.product(dtype=[jnp.float32, jnp.bfloat16], )
+    def test_ragged_paged_attention_decode_only(self, dtype):
+        seq_lens = [
+            (1, 18),
+            (1, 129),
+            (1, 597),
+            (1, 122),
+            (1, 64),
+            (1, 322),
+            (1, 463),
+            (1, 181),
+            (1, 1107),
+            (1, 123),
+            (1, 31),
+            (1, 18),
+            (1, 1229),
+            (1, 229),
+            (1, 87),
+            (1, 1328),
+        ]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            dtype,
+            dtype,
+            num_pages,
+        )
+
+    @parameterized.product(dtype=[jnp.float32, jnp.bfloat16], )
+    def test_ragged_paged_attention_prefill_only(self, dtype):
+        seq_lens = [
+            (5, 18),
+            (15, 129),
+            (120, 597),
+            (100, 122),
+            (21, 64),
+            (32, 322),
+            (251, 463),
+            (40, 181),
+            (64, 1107),
+            (99, 123),
+            (10, 31),
+            (5, 18),
+            (3, 1229),
+            (120, 229),
+            (9, 87),
+            (2, 1328),
+        ]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            dtype,
+            dtype,
+            num_pages,
+        )
+
+    @parameterized.product(dtype=[jnp.float32, jnp.bfloat16], )
+    def test_ragged_paged_attention_mixed(self, dtype):
+        seq_lens = [
+            (5, 18),
+            (1, 129),
+            (120, 597),
+            (1, 122),
+            (1, 64),
+            (32, 322),
+            (251, 463),
+            (1, 181),
+            (1, 1107),
+            (99, 123),
+            (1, 31),
+            (5, 18),
+            (3, 1229),
+            (117, 229),
+            (1, 87),
+            (1, 1328),
+        ]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            dtype,
+            dtype,
+            num_pages,
+        )
+
+    @parameterized.product(
+        num_seqs=[1, 17],
+        num_heads=[(32, 8), (12, 2), (5, 1), (3, 3)],
+        head_dim=[80, 240],
+        dtype=[jnp.float32, jnp.bfloat16],
+        # num_kv_pages_per_block=[8, 16],
+        # num_queries_per_block=[16, 32],
+    )
+    def test_ragged_paged_attention_complex(
+        self,
+        num_seqs,
+        num_heads,
+        head_dim,
+        dtype,
+        # num_kv_pages_per_block,
+        # num_queries_per_block,
+    ):
+        rng = np.random.default_rng(1234)
+        q_lens = rng.integers(1, 100, num_seqs)
+        kv_lens = q_lens + rng.integers(0, 50, num_seqs)
+        seq_lens = list(zip(q_lens.tolist(), kv_lens.tolist()))
+        page_size = 16
+        num_pages = 1000
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            dtype,
+            dtype,
+            num_pages,
+            # num_kv_pages_per_block=num_kv_pages_per_block,
+            # num_queries_per_block=num_queries_per_block,
+        )
+
+    @parameterized.product(sliding_window=[None, 5, 128], )
+    def test_ragged_paged_attention_sliding_window(
+        self,
+        sliding_window: int | None,
+    ):
+        num_seqs = 5
+        num_heads = (4, 4)
+        dtype = jnp.float32
+        rng = np.random.default_rng(1234)
+        q_lens = rng.integers(1, 100, num_seqs)
+        kv_lens = q_lens + rng.integers(0, 50, num_seqs)
+        seq_lens = list(zip(q_lens.tolist(), kv_lens.tolist()))
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            dtype,
+            dtype,
+            num_pages,
+            sliding_window=sliding_window,
+        )
+
+    @parameterized.product(soft_cap=[None, 50.0], )
+    def test_ragged_paged_attention_logit_soft_capping(
+        self,
+        soft_cap: float | None,
+    ):
+        num_heads = (16, 2)
+        num_seqs = 2
+        dtype = jnp.float32
+        rng = np.random.default_rng(1234)
+        q_lens = rng.integers(1, 100, num_seqs)
+        kv_lens = q_lens + rng.integers(0, 50, num_seqs)
+        seq_lens = list(zip(q_lens.tolist(), kv_lens.tolist()))
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        self._test_ragged_paged_attention(
+            seq_lens,
+            num_heads,
+            head_dim,
+            page_size,
+            dtype,
+            dtype,
+            num_pages,
+            soft_cap=soft_cap,
+        )
+
+    def test_ragged_paged_attention_sliding_window_should_be_positive(self):
+        dtype = jnp.float32
+        seq_lens = [(192, 328), (128, 180), (64, 255)]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        with self.assertRaisesRegex(ValueError, "must be positive"):
+            self._test_ragged_paged_attention(
+                seq_lens,
+                num_heads,
+                head_dim,
+                page_size,
+                dtype,
+                dtype,
+                num_pages,
+                sliding_window=0,
+            )
+
+        with self.assertRaisesRegex(ValueError, "must be positive"):
+            self._test_ragged_paged_attention(
+                seq_lens,
+                num_heads,
+                head_dim,
+                page_size,
+                dtype,
+                dtype,
+                num_pages,
+                sliding_window=-1,
+            )
+
+    def test_ragged_paged_attention_soft_cap_cannot_be_zero(self):
+        dtype = jnp.float32
+        seq_lens = [(192, 328), (128, 180), (64, 255)]
+        num_heads = (32, 8)
+        head_dim = 128
+        page_size = 16
+        num_pages = 1000
+
+        with self.assertRaisesRegex(ValueError, "must not be 0.0"):
+            self._test_ragged_paged_attention(
+                seq_lens,
+                num_heads,
+                head_dim,
+                page_size,
+                dtype,
+                dtype,
+                num_pages,
+                soft_cap=0.0,
+            )
+
+
+if __name__ == "__main__":
+    absltest.main(testLoader=jtu.JaxTestLoader())

tests/lora/test_lora.py

@@ -0,0 +1,123 @@
+# https://github.com/vllm-project/vllm/blob/ed10f3cea199a7a1f3532fbe367f5c5479a6cae9/tests/tpu/lora/test_lora.py
+import pytest
+import vllm
+from vllm.lora.request import LoRARequest
+
+# This file contains tests to ensure that LoRA works correctly on the TPU
+# backend. We use a series of custom trained adapters for Qwen2.5-3B-Instruct
+# for this. The adapters are:
+# Username6568/Qwen2.5-3B-Instruct-1_plus_1_equals_x_adapter, where x ranges
+# from 1 to 4.
+
+# These adapters are trained using a standard huggingface peft training script,
+# where all the inputs are "What is 1+1? \n" and all the outputs are "x". We run
+# 100 training iterations with a training batch size of 100.
+
+
+@pytest.fixture(scope="function", autouse=True)
+def use_v1_only(monkeypatch: pytest.MonkeyPatch):
+    """
+    Since Multi-LoRA is only supported on the v1 TPU backend, set VLLM_USE_V1=1
+    for all tests in this file
+    """
+    with monkeypatch.context() as m:
+        m.setenv("VLLM_USE_V1", "1")
+        yield
+
+
+def setup_vllm(num_loras: int) -> vllm.LLM:
+    return vllm.LLM(model="Qwen/Qwen2.5-3B-Instruct",
+                    max_model_len=256,
+                    max_num_seqs=8,
+                    enable_lora=True,
+                    max_loras=num_loras,
+                    max_lora_rank=8)
+
+
+def test_single_lora():
+    """
+    This test ensures we can run a single LoRA adapter on the TPU backend.
+    We run "Username6568/Qwen2.5-3B-Instruct-1_plus_1_equals_2_adapter" which
+    will force Qwen2.5-3B-Instruct to claim 1+1=2.
+    """
+
+    llm = setup_vllm(1)
+
+    prompt = "What is 1+1? \n"
+
+    lora_request = LoRARequest(
+        "lora_adapter_2", 2,
+        "Username6568/Qwen2.5-3B-Instruct-1_plus_1_equals_2_adapter")
+    output = llm.generate(prompt,
+                          sampling_params=vllm.SamplingParams(max_tokens=16,
+                                                              temperature=0),
+                          lora_request=lora_request)[0].outputs[0].text
+
+    answer = output.strip()[0]
+
+    assert answer.isdigit()
+    assert int(answer) == 2
+
+
+def test_lora_hotswapping():
+    """
+    This test ensures we can run multiple LoRA adapters on the TPU backend, even
+    if we only have space to store 1.
+
+    We run "Username6568/Qwen2.5-3B-Instruct-1_plus_1_equals_x_adapter" which
+    will force Qwen2.5-3B-Instruct to claim 1+1=x, for a range of x.
+    """
+
+    lora_name_template = \
+        "Username6568/Qwen2.5-3B-Instruct-1_plus_1_equals_{}_adapter"
+    lora_requests = [
+        LoRARequest(f"lora_adapter_{i}", i, lora_name_template.format(i))
+        for i in range(1, 5)
+    ]
+
+    llm = setup_vllm(1)
+
+    prompt = "What is 1+1? \n"
+
+    for i, req in enumerate(lora_requests):
+        output = llm.generate(prompt,
+                              sampling_params=vllm.SamplingParams(
+                                  max_tokens=16, temperature=0),
+                              lora_request=req)[0].outputs[0].text
+        answer = output.strip()[0]
+
+        assert answer.isdigit()
+        assert int(answer) == i + 1, f"Expected {i + 1}, got {answer}"
+
+
+def test_multi_lora():
+    """
+    This test ensures we can run multiple LoRA adapters on the TPU backend, when
+    we have enough space to store all of them.
+
+    We run "Username6568/Qwen2.5-3B-Instruct-1_plus_1_equals_x_adapter" which
+    will force Qwen2.5-3B-Instruct to claim 1+1=x, for a range of x.
+    """
+    lora_name_template = \
+        "Username6568/Qwen2.5-3B-Instruct-1_plus_1_equals_{}_adapter"
+    lora_requests = [
+        LoRARequest(f"lora_adapter_{i}", i, lora_name_template.format(i))
+        for i in range(1, 5)
+    ]
+
+    llm = setup_vllm(4)
+
+    prompt = "What is 1+1? \n"
+
+    for i, req in enumerate(lora_requests):
+        output = llm.generate(prompt,
+                              sampling_params=vllm.SamplingParams(
+                                  max_tokens=16, temperature=0),
+                              lora_request=req)[0].outputs[0].text
+
+        answer = output.strip()[0]
+
+        assert answer.isdigit()
+        assert int(
+            output.strip()
+            [0]) == i + 1, f"Expected {i + 1}, got {int(output.strip()[0])}"